Attackers can elicit 'toxic behavior' from AI translation systems, study finds

In a new study, researchers demonstrate a flaw in neural machine translation systems that can prompt them to generate toxic outputs.